One of those things I thought would be easy turned out to be incredibly hard.
Labs & Measurement is the keystone entrypoint for data captured from doctors’ notes, lab reports, and devices. Building it required a massive extension of our underlying ontology. We had to learn how to express bounded values like <2 while still enforcing validation that values remain greater than zero.
MCP kicked into high gear. Slida’s internal agent can now produce a copy-and-paste prompt for external MCP setup. This let us put frontier agents to work ahead of purpose-trained vision, OCR, and PDF handling.
I threw an entire MyChart export at it and, well, it failed. Turns out we measure a lot. It also turns out a lot of important information in those reports is not a measurement. The community of providers is sometimes more important than the value itself—and in some cases, the value is useless without knowing who ordered it.
Export to FHIR was supposed to be one of the easy ones too. Turns out it isn’t. There is no single published FHIR string-to-type mapping we can use for this translation, and we found a lot of edge cases.
It is fun to search labs by name and see values in near real time. I am also looking forward to a nice 0.2.7. We have milliseconds to shave because we are re-projecting half the database on every click. It is kind of amazing that we are only seeing 0.5 ms stutters when nothing is cached.
Having the data locally is a completely different performance regime. We can control display timing and page draws like a game.
Hiring UX soon…